Surgical endoscopy · 2026
Multimodal models based on radiomics and deep learning in the classification of gastric stromal tumors and gastric leiomyomas.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Xu Guoting, Zhou Ye, Liu Lihe, Zou Zhaoyang, Zhu Zixiang, Zhu Jinzhou, Wu Airong
Аннотация
OBJECTIVE: To establish and validate a multimodal model based on radiomics and deep learning for classifying gastric stromal tumors (GSTs) and gastric leiomyomas (GLMs).
METHODS: The study collected clinical case data of 200 patients (137 GSTs and 63 GLMs) hospitalized for gastric submucosal tumors at the First Affiliated Hospital of Soochow University. The collected data included white light endoscopy, ultrasound endoscopic images, computed tomography (CT) images, records of endoscopic intervention, and pathological results. The patients were randomly divided into a training set (134 cases) and a validation set (66 cases) at a ratio of 7:3. A radiomics model was constructed by extracting radiomic features from three-dimensional CT images using XGBoost algorithm, while a deep learning model was developed based on convolutional neural network to extract deep features. Finally, clinical features and the predictions from the models were integrated to establish a multimodal model based on XGBoost algorithm. To enhance model visualization, variable importance ranking and local interpretable visualization were used. Area under the curve (AUC), sensitivity, specificity, false-negative rate (FNR), decision curve analysis net benefit, calibration, and tenfold stratified cross-validation stability assessed model performance.
RESULTS: In the multimodal model for the classification of GSTs and GLMs, the prediction of the deep learning model was the most important variable, followed by tumor echo homogeneity, age, the prediction of the radiomics model, and major axis of tumor. The multimodal model achieved excellent AUCs (0.927 in the training set, 0.882 in the validation set), outperforming both radiomics and deep learning models. It exhibited favorable calibration, minimal overfitting, a low FNR, and stable performance via tenfold cross-validation.
CONCLUSION: Based on the integration of clinical tabular data, radiomics, and deep learning features, the multimodal model could classify GSTs and GLMs, whose performance was better than the single modalities.
METHODS: The study collected clinical case data of 200 patients (137 GSTs and 63 GLMs) hospitalized for gastric submucosal tumors at the First Affiliated Hospital of Soochow University. The collected data included white light endoscopy, ultrasound endoscopic images, computed tomography (CT) images, records of endoscopic intervention, and pathological results. The patients were randomly divided into a training set (134 cases) and a validation set (66 cases) at a ratio of 7:3. A radiomics model was constructed by extracting radiomic features from three-dimensional CT images using XGBoost algorithm, while a deep learning model was developed based on convolutional neural network to extract deep features. Finally, clinical features and the predictions from the models were integrated to establish a multimodal model based on XGBoost algorithm. To enhance model visualization, variable importance ranking and local interpretable visualization were used. Area under the curve (AUC), sensitivity, specificity, false-negative rate (FNR), decision curve analysis net benefit, calibration, and tenfold stratified cross-validation stability assessed model performance.
RESULTS: In the multimodal model for the classification of GSTs and GLMs, the prediction of the deep learning model was the most important variable, followed by tumor echo homogeneity, age, the prediction of the radiomics model, and major axis of tumor. The multimodal model achieved excellent AUCs (0.927 in the training set, 0.882 in the validation set), outperforming both radiomics and deep learning models. It exhibited favorable calibration, minimal overfitting, a low FNR, and stable performance via tenfold cross-validation.
CONCLUSION: Based on the integration of clinical tabular data, radiomics, and deep learning features, the multimodal model could classify GSTs and GLMs, whose performance was better than the single modalities.